Aholo 3DGS Reconstruction and Generation

Problem It Solves

When a user asks for “3D from video/images,” the task is not 2D image generation. It is creating Aholo 3DGS reconstruction or generation jobs, tracking a worldId, and retrieving model artifacts such as PLY or SPZ. This skill turns the 2D/3D ambiguity, parameter confirmation, upload, creation, polling, and error handling into one Agent-executable workflow.

How It Works

• Scope: handles 3D reconstruction, generation, worldId lookup, and artifact retrieval; does not handle 2D text-to-image.
• Execution: the Agent normally runs aholo_reconstruct.py with AHOLO_API_KEY configured; the flow gets an upload token, uploads assets, calls /world/v1/*, and records worldId.
• Reconstruction checks: before creation, it confirms scene, taskQuality, and useMask in one turn; useMask applies only when scene=model, preventing silent defaults.
• Cost control: one create call is allowed per intent by default; for multiple videos, the user must choose whether to merge them into one 3DGS or create separate jobs, avoiding duplicate charges.
• Result follow-up: after creation, it asks whether to wait, then either polls synchronously or returns worldId and the Studio link; failures are diagnosed from bizCode values such as unauthenticated or insufficient credits.

Boundaries and Notes

Use it only for explicit 3D tasks, not requests like “generate one render.” If an image directory is provided, the whole directory must be uploaded. For private TLS certificates, prefer configuring the CA bundle rather than disabling certificate verification.